MAINT: optimize.root_scalar: ensure that root is a scalar

[mdhaber]

Reference issue

gh-17691

What does this implement/fix?

gh-17691 removed the restriction to use scipy.optimize.root_scalar with method='newton' without fprime; the derivative is calculated numerically by approx_fprime. However, approx_fprime returns an array even with scalar input, so root_scalar would end up returning an array with a single element rather than a scalar. This fixes the inconsistency by ensuring that the scalar derivative is extracted from the output of approx_fprime.

Additional information

import scipy as sp

def f(x):
    return (x - 1) * (x - 2)

res = sp.optimize.root_scalar(f, x0=0)
res.root.shape  # was (1,), now ()

[tupui]

LGTM, merging once green :)

[j-bowhay]

Thanks @mdhaber